at://
/
app.bsky.feed.post
/
3mkzsnojtxt22
sign in
All
4
Record
2
Post
1
PostEmbed
1
Post
by @danabra.mov
PostEmbed
by @danabra.mov
Record
by @jimpick.com
Record
by @atsui.org
+ new component
Post
To study this, we build FINER, with large-scale benchmarks: FINER-CompreCap and FINER-DOCCI. We construct paired positive/negative MCQs by composing objects, attributes, and relations, then changing one fine-grained detail at a time.
1mo
ExplainableML
